A social recommender system based on exponential random graph model and sentiment similarity

With the rapid growth of social media, recommendation for social activities is urgently needed to overcome information overload. Micro-blog, as one of most popular social media platform, has not a good enough recommender approach to satisfy users' expectation. In this paper, we proposed a social recommender system using both exponential random graph model and sentiment similarity. Firstly, we built a good fitted graph model that was used to predict the probabilities of non-linked nodes. Moreover, we collected contents of each user for mining their emotions and select 106 features. Karhunen-Loeve transform (KLT) was chose to analyze the features of those texts. Based on KLT, average distances of text features were used to calculate the sentiment similarity. Therefore, according to the resort of similarities, we gave top-N recommendation for user. Finally, we studied this social recommender system on diabetes micro-blog. The metrics showed that our proposed social recommender system outperform other methods.